Most lead qualification systems ask the same questions in the same order forever. Leads.cx uses an AI layer that analyses what happened in thousands of qualification calls — comparing screening responses against actual close rates — and continuously adjusts which signals predict a winning lead. The result compounds over time: the longer a subscriber has been on the platform, the better the quality of their leads.
+22%
Close Rate Improvement
Weekly
Model Refinement Cadence
6 mo
Time to Full Improvement
The Problem With Static Qualification
Every lead qualification system starts with a fixed set of criteria: intent, service area, budget, timing. These criteria are defined at launch based on available knowledge — what the industry generally treats as a qualified lead — and they work reasonably well. But they have a structural limitation: they don't learn from outcomes.
A static qualification model treats every lead who passes the standard screening as equally likely to close. In practice, that's never true. Some combinations of answers predict a winning lead far better than others. A homeowner who says they have storm damage, are the primary decision-maker, and are looking to start in the next two weeks closes at a very different rate than one who says they're "thinking about it" and "checking prices." Both pass the basic criteria. One converts at 55%. The other at 18%.
Without a mechanism to detect and act on those patterns, a qualification system delivers leads of uneven quality even when all of them pass the standard criteria. And it can't tell the difference.
The other issue is that markets change. What predicted a high-intent buyer in 2022 may be less predictive in 2025. Seasonal behaviour shifts. Local economic conditions evolve. Consumer communication patterns change. A qualification model that doesn't update based on actual outcome data falls behind reality over time.
"We'd been getting decent leads for about eight months. The close rate was good. Then in month nine, something changed — it was like the leads got sharper. We were closing a noticeably higher percentage. When we asked what happened, they explained that the AI model had identified some new patterns in our category and updated the qualification criteria. We hadn't changed anything on our end. The product just got better."
How It Works
The AI qualification layer sits between the qualification call data and the delivery decision. It doesn't replace the qualification criteria — it evaluates each lead against a scoring model that weights criteria based on their historical correlation with close rate in that specific industry and geography.
Every delivered lead generates two data points over time: what was said in the qualification call (the inputs) and whether the lead converted into a booked job (the outcome). Subscriber close rate reporting — submitted through regular feedback — creates the outcome side of the loop.
The AI model analyses these paired data sets to identify which combinations of qualification signals are most predictive of a closed job in each category. It looks for patterns that wouldn't be visible in manual review: correlations between specific phrasing in the call, answer combinations, response timing, and close rate across thousands of leads.
Leads that score below the threshold for a given subscriber's category and geography are discarded — even if they pass the base criteria. This is what produces the continuous quality improvement that subscribers experience over time.
The model is updated on a weekly cadence rather than through periodic manual reviews. This means the qualification criteria for a roofing subscriber in a hail-prone market in May will reflect the current pattern of high-intent storm leads, not the same model that was in place during a quiet winter period.
Seasonal shifts in consumer behaviour, changes in local market conditions, and evolving conversion patterns are captured in the weekly data cycle and reflected in the scoring model before they would be visible to a human reviewer looking at monthly reports.
This means that a subscriber's lead quality doesn't plateau. It continues to improve as the model accumulates outcome data specific to their category, geography, and subscriber feedback.
In the first month of a subscription, lead quality reflects the baseline qualification criteria for the category. These are good — they reliably filter out leads that clearly won't convert. By month three, the AI layer has enough outcome data to begin adjusting the scoring thresholds. By month six, the model has identified category-specific patterns that were invisible at launch.
Across subscribers tracked over a 6-month window, the average close rate improvement between month one and month six was 22%. This improvement didn't require any changes on the subscriber's end — it was entirely driven by the qualification model getting more accurate over time.
How the System Works in Practice
The qualification system operates in three layers. Each builds on the previous and together they create a lead delivery model that improves continuously rather than requiring periodic overhauls.
The standard qualification call covers four criteria: confirmed service need, confirmed service area, confirmed decision-maker, confirmed 30-day intent. Any lead that doesn't clear all four is discarded regardless of scoring. This is the floor — no lead is delivered without passing these minimum criteria.
The base criteria are category-specific and reviewed quarterly. A plumbing lead and a cosmetic dentistry lead have different base criteria because the nature of urgency, budget, and decision-making is different in each case.
Leads that pass the base criteria are scored against the AI model for their specific category and geography. The score reflects the probability of close based on the pattern of their qualification call answers compared to historical outcome data. Leads below the score threshold are discarded even though they passed the base criteria.
The scoring threshold adjusts weekly based on incoming outcome data. When the model identifies a new pattern that correlates with conversion — for example, that leads mentioning a specific trigger event in a given category close at twice the rate of the base pool — that signal is weighted more heavily in subsequent scoring.
Subscribers report close rate outcomes through a simple monthly check-in — which leads converted and which didn't. This outcome data is the raw material that trains the AI model. Without it, the model can only use general category patterns. With it, it adjusts to the specific close profile of each subscriber's business.
Subscribers who provide consistent outcome feedback receive leads that are calibrated specifically to their close profile within 90 days. The model learns what their business wins and prioritises those patterns in subsequent delivery.
The Results
+22%
Close Rate Improvement
Average improvement across subscribers tracked from month one to month six. Entirely driven by AI model improvement — no change required from the subscriber.
Weekly
Model Refinement Cadence
Qualification scoring thresholds update weekly based on outcome data — capturing seasonal shifts and market changes as they happen.
6 mo
Time to Full Improvement
The model accumulates enough category-specific outcome data to reach full refinement by month six for most service categories.
Feedback accelerates the improvement significantly. Subscribers who provide monthly close rate reports get leads calibrated to their specific business profile within 90 days. Subscribers who don't report outcomes still benefit from category-level model updates — which improve for the whole category as outcomes are reported across all subscribers in that category — but the improvement is less specific to their close profile.
A close is a lead that converted into a booked and completed job — or, for ongoing service contracts, a signed agreement. Leads that you contacted but couldn't close are also valuable feedback: knowing which lead profile consistently doesn't convert helps the model reduce those profiles in future delivery. You report two pieces of data: lead ID and outcome (closed / not closed). Optional: reason if not closed.
Yes — category-level patterns are derived from aggregate data across all subscribers in that category. Your individual close rate data is never shared with or visible to other subscribers. The model uses your outcomes to calibrate your specific profile, and uses category-aggregate data to identify broader patterns (seasonal signals, geography effects, phrasing correlations) that improve the base model for the whole category.
Most lead scoring systems score at the time of inquiry based on form data — demographics, zip code, form field completion. They don't update based on outcome data, don't incorporate what was actually said in a live qualification call, and don't improve over time. The Leads.cx model scores based on the full qualification call transcript plus historical close rate data for that answer pattern in your specific category and geography. The outcome feedback loop is what makes it improve rather than plateau.
The subscription guarantee is for 30 qualified leads per month — that number is maintained. What changes as the model improves is the quality floor within that 30, not the count. Leads that would have cleared the base criteria but scored low on the AI model are discarded and replaced with leads that score higher. The number stays constant; the average quality of the 30 increases.
The longer you're on the platform, the better the leads get. Start the feedback loop and let the qualification model learn what your business wins.